Smarter Backoff Waits

[tmagrino]

Right now there's a variety of reasons a transaction can roll back which do not really merit waiting for exponential backoff, eg. abort due to an update to the worker cache.

We should change the backoff procedure to reset/skip the delay depending on the reason for the transaction's abort. I expect there's a nuanced and deep design space here that could be interesting to explore and perhaps write a paper about?

[tmagrino]

We currently have @samanthad98 looking into this as an undergrad research task.